Unable to use model trained from enhancement template
Zahuczky opened this issue · 1 comments
Zahuczky commented
Describe the bug
I trained with my custom dataset with minimal modifications to https://github.com/speechbrain/speechbrain/tree/develop/templates/enhancement (0.5.16 version)
After, I end up with a result folder structure such as
results
└───4234
│ env.log
│ hyperparams.yaml
│ log.txt
│ train.py
│ train_log.txt
│
└───save
└───CKPT+2024-02-15+17-49-49+00
brain.ckpt
CKPT.yaml
counter.ckpt
dataloader-TRAIN.ckpt
model.ckpt
optimizer.ckpt
How could I load/use the model I trained for enhancement in my python script, like
model.enhance_file("example.wav", "output.wav")
Expected behaviour
Being able to use it like
model.enhance_file("example.wav", "output.wav")
To Reproduce
No response
Environment Details
No response
Relevant Log Output
No response
Additional Context
No response
asumagic commented
Currently, the process is manual. Typically, you would:
- Look up a relevant interface in
speechbrain/inference
(as of 1.0), e.g.WaveformEnhancement
- Check what hparams/mods it attempts to use
- Copy over the relevant files in a directory (e.g. here likely just
model.ckpt
) - Build an inference
hyperparams.yaml
in the same directory from your training hyperparams (optionally taking inspiration from a similar published model on HF) - Finally, write and use a test script that uses the inference interface e.g.
enhancement = WaveformEnhancement.from_hparams("./mymodeldirectory/", run_opts={"device": device})